As a voice recognition method, a method of recognizing a voice uttered after a speaking person's operation of commanding a voice recognition start and a method of recognizing an uttered voice at all times without such an operation are known widely. A problem with the former method is that while the recognition rate becomes high because the starting end of an uttered voice is specified and therefore the detection of a section is facilitated, there is a necessity to perform a specific operation every time when recognition is performed and hence an inconvenience is imposed on the speaking person. On the other hand, a problem with the latter method is that because no specific operation is needed, the above-mentioned disadvantage is eliminated, but the recognition rate becomes low because it is difficult to detect a voice section.
Therefore, in order to solve these problems, for example, patent reference 1 discloses a voice recognition device that is provided with the above-mentioned two voice recognition methods, and that automatically switches to an appropriate one of the voice recognition methods according to a speaking person's state or a surrounding state. Concretely, when, for example, the surrounding noise is large, the voice recognition device switches to the former voice recognition method, whereas when the noise is small, switches to the latter voice recognition method.